KR20080024776A - Method and apparatus for recognizing parking slot marking by using hough transformation and parking assist system using same - Google Patents

Abstract

A method and an apparatus for recognizing a parking lot, and a parking assistance system using the apparatus are provided to allow a driver to set a desired parking position accurately, by using an image of a camera to recognize a parking lot within a vehicle. An input image is received from a camera, and then converted into an edge image of an air view. A Hough space image is formed by performing Hough transformation on the edge image of the air view(S220). Parking lot-marking lines are recognized on the edge image of the air view by detecting a plurality of peak pairs on the Hough space image(S230). A plurality of parking lot-marking line segments is recognized from the parking lot-marking lines(S240). A guide line is recognized from the parking lot-marking line segments(S250). A plurality of divisional marking line segments is recognized from the guide line(S260). A parking lot is recognized by the guide line and divisional marking line segments(S270).

Description

Method and Apparatus for Recognizing Parking Lot Using Hough Transformation and Method for Apparatus and Recognizing Parking Slot Marking by Using Hough Transformation and Parking Assist System Using Same}

1 is a block diagram schematically showing a parking assist system using a hop transform according to an embodiment of the present invention;

2 is a flowchart illustrating a parking block recognition method using a hop transform according to an embodiment of the present invention;

3 is an exemplary diagram for explaining three coordinate systems;

4 is an exemplary diagram for explaining a relationship between a D coordinate system and a U coordinate system;

5 is an exemplary diagram for explaining a conversion process between a U coordinate system and a B coordinate system;

6 is an exemplary diagram for explaining a process of converting a pixel coordinate system into an image index coordinate system;

7 is an exemplary view for explaining a process of converting from a world coordinate system to a bird's eye view index coordinate system;

8 is an exemplary view for explaining a view into a bird's eye view and a bird's eye view corner image;

9 is an exemplary diagram for explaining that a point on a bird's eye view corner image is converted into a hop space image;

10 is an exemplary diagram showing a straight line on a bird's eye view corner image as a plurality of curves on a hop space image;

11 is an exemplary view showing two parallel straight lines on a bird's eye view corner image on a hop space image;

12 is an exemplary view showing several parallel straight lines of a parking lot bird's eye view on a hop space image;

13 is an exemplary view for explaining a peak pair detection process according to a preferred embodiment of the present invention;

14 is an exemplary view for explaining a process of extracting a parking section display line by detecting a cluster;

15 is an exemplary view for explaining a process of obtaining an end point of a parking segment display line in a bird's eye view corner image;

16 is a flowchart for explaining a process of recognizing parking segment display line segments;

17 is an exemplary view showing a recognition result of a parking segment display line on a bird's eye view;

18 is an exemplary diagram illustrating a vector for recognizing a guideline;

19 is an exemplary diagram for explaining a process of stably detecting a guideline regardless of a detection position of an end point of a display line segment;

20 is an exemplary view illustrating a result of recognizing a guideline according to a preferred embodiment of the present invention;

21 is an exemplary view for explaining a process of recognizing a segment display line segment according to a preferred embodiment of the present invention;

22 is an exemplary view showing a guideline extension line and I _ON and I _OFF of a bird's eye view;

23 is an exemplary diagram showing L _dividing (s) and a threshold;

24 is an exemplary view showing a result of recognizing division display lines;

25 is an exemplary view illustrating a result of recognizing a parking compartment according to a parking compartment recognition method according to a preferred embodiment of the present invention;

FIG. 26 is an exemplary diagram illustrating a process of recognizing segment display line segments in backlight; FIG.

27 is an exemplary diagram illustrating a process of recognizing a segment display line segment when light is reflected.

<Description of Symbols for Main Parts of Drawings>

100: parking assistance system 110: target parking position setting unit

112: camera 114: parking compartment recognition unit

116: user interface unit 120: detection unit

130: parking assistance control unit 132: route plan generation unit

134: route tracking industry 136: position estimation unit

140: active steering

The present invention relates to a parking compartment recognition method using Hough Transformation, a device and a parking assistance system using the same. More specifically, in recognizing a parking section by photographing a parking space in a vehicle, a parking section display line is recognized by detecting a peak pair having the same direction at a predetermined distance away from the hop space by using a hop transform. The parking segment indicator is recognized by inspecting the presence or absence of corner pixels at the center and boundary along the display line, and the parking segment indicator segment closer to the camera and more perpendicular to the camera's line of sight as a guideline among the parking segment indicator segments. In addition, the present invention relates to a method, apparatus, and a parking assistance system using the same to recognize parking segment more accurately and quickly by performing contrast contrast according to guidelines.

Typically, drivers who are inexperienced in driving such as women and the elderly have a lot of difficulties in parking backwards between the vehicles parked in the parking lot. To this end, most vehicles employ ultrasonic parking assistance systems to alert drivers of stopping distances so that inexperienced drivers can avoid many obstacles, such as parked vehicles, people and arrangements.

For example, certain vehicles are equipped with a rear guide monitor to assist the driver's parking by examining a driving route predicted by the image of the rear camera. Recent vehicles also have environmental awareness in parking assistance systems to provide drivers with better rear vision. Vehicles equipped with these parking assistance systems use techniques such as Wheel Speed Sensor and Intermediate View Reconstruction (IVR) to create virtual images from the optimal view point. use.

Meanwhile, an automatic parking system is being developed by further modifying the parking assistance system. Automatic parking systems automate driving for parking with automatic steering control and automatic braking control. The automatic parking system includes a path generation device for measuring the position of the target location, an automatic steering / braking system for driving the vehicle in a predicted path, a user interface for receiving visual information about the vehicle's progress based on a driver's input, and the like. Equipped.

In order to perform automatic parking or semi-automatic parking using the automatic parking system in the vehicle, it is essential to accurately recognize the target position. As a method of recognizing the target position, a method by manual instruction by a driver, a method using a global positioning system (GPS), a method of recognizing a vision-based parking compartment using a camera, and the like are used.

Here, the manual instruction by the driver requires the driver to set up and recognize the parking section every time, and the procedure is complicated and cumbersome. The method using the satellite navigation system increases the cost of purchasing the satellite navigation device. Considering that parking is performed in a small space, there is a disadvantage in that the utilization is inferior, and a vision based parking compartment recognition method using a camera is often used.

Therefore, in utilizing the vision-based parking compartment recognition method, a method for recognizing the parking compartment more precisely, quickly and simply is required.

In order to meet this demand, the present invention uses a hop transform to detect a parking section by capturing a parking space in a vehicle and detecting peak pairs having the same direction at a predetermined distance away from the hop space. Recognize the parking segment by checking the presence or absence of corner pixels at the center and boundary along the parking segment indicator line, and identify the segment of parking segment that is closer to the camera and more perpendicular to the direction of the camera. The object of the present invention is to provide a method, apparatus, and a parking assistance system using the same, which are recognized as a guideline, by performing contrast contrast according to the guideline, and recognizing the segmentation line segment, to more accurately and quickly recognize the parking compartment.

In order to achieve the above object, the present invention, the parking space in the vehicle to generate an input image after the camera and the driver receives a reference point of the parking space from the driver to receive and deliver the parking section to recognize the parking compartment In the device, when the input image is received from the camera, the input image is converted into a bird's eye view corner image, Hough transforms the bird's eye view corner image to generate a hop space image, and then separated by a predetermined distance on the hop space image. By detecting peak pairs with directions, the parking lot indicator lines are recognized on the bird's eye view corner image, the parking lot indicator lines are recognized by the parking lot indicator lines, and the guideline is used by the parking lot indicator lines. After recognizing the It provides a parking compartment recognition apparatus using a hop transform characterized in that it comprises a parking section recognition unit for recognizing a plurality of division display line segments using a guide line and a plurality of division display line segments.

In addition, according to another object of the present invention, the parking compartment is connected to the user interface to receive and transmit a reference point to be parked in the parking space from the camera and the driver to shoot the parking space in the vehicle to generate an input image and then transfer A method for recognizing a parking compartment in a parking compartment recognizing unit, the method comprising: (a) generating a bird's eye view corner image by receiving an input image, and hop-converting the bird's eye view corner image to generate a hop space image; (b) detecting a plurality of peak pairs having the same direction at a predetermined distance apart from the hop spatial image; (c) converting the hop spatial image into a bird's eye view corner image to recognize the plurality of peak pairs as a plurality of parking section indicator lines; (d) recognizing a plurality of parking section display lines from the plurality of parking section display lines; (e) recognizing a guideline from the plurality of parking compartment marking segments; And (f) recognizing a plurality of segment display line segments using the guideline.

In addition, according to another object of the present invention, in the parking assistance system for assisting the parking of the vehicle by recognizing the parking compartment, when the input image is received from the camera, the input image is converted into a bird's eye view corner image, the bird's eye view corner image After Hough transforming to generate a hop space image, a plurality of parking zone indicator lines are recognized on the bird's eye view corner image by detecting peak pairs having the same direction at a predetermined distance from the hop space image. After extracting the guideline and the number of division mark segments from the parking lot indicator lines, it recognizes them as parking compartments, converts the bird's eye view corner image into the output image, and outputs it. After receiving the reference point from the driver, the reference point is projected to the guideline. Target parking by setting and passing the target parking compartment Value setting unit; A sensing unit having a plurality of sensors for recognizing a driving state of the vehicle to generate and transmit driving state information of the vehicle; Receives information on the driving situation from the sensor to estimate the location of the vehicle, receives the target parking block, generates a route plan for parking in the parking compartment, and generates a control signal for parking in the parking compartment in consideration of the location. A parking auxiliary control unit for transmitting; And an active steering unit for driving the vehicle in response to the control signal when receiving the control signal.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. First of all, in adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are used as much as possible even if displayed on different drawings. In addition, in describing the present invention, when it is determined that the detailed description of the related well-known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

1 is a block diagram schematically showing a parking assist system using a hop transform according to a preferred embodiment of the present invention.

The parking assistance system 100 using the hop transform according to the preferred embodiment of the present invention includes a target parking position setting unit 110, a sensing unit 120, a parking assistance control unit 130, and an active steering unit 140. .

The target parking position setting unit 110 includes a camera 112, a parking compartment recognition unit 114, and a user interface unit 116 to recognize a parking compartment from an input image of the parking space photographed by the camera 112. It is a means for setting the parking position that the driver wants to park.

The camera 112 according to an exemplary embodiment of the present invention generates and transmits an input image by photographing a parking space.

The parking compartment recognition unit 114 according to the preferred embodiment of the present invention converts the input image into a bird's eye view corner image, and hop-converts the bird's eye view corner image to generate a hop space image, and the same direction is separated by a predetermined distance from the hop space image. Recognizes a number of parking segment display lines on the bird's eye view corner image by detecting peak pairs, and recognizes a plurality of parking segment display lines on a plurality of parking segment display lines, and recognizes guidelines on a plurality of segmentation lines. Recognize the segmentation line segment using the guideline, recognize the segment line segment and the segmentation line segment as a parking section, convert the bird's eye view corner image into the output image, transfer the image to the user interface unit 116, and then the user interface unit 116 ), Then pass the reference point to the guidelines Projection sets the target parking compartment and transfers it to the parking assistance controller 130.

To this end, the parking compartment recognition unit 114 according to a preferred embodiment of the present invention includes a memory for storing software and data of an algorithm for recognizing a parking compartment and a microprocessor for recognizing the parking compartment by executing a program stored in the memory. It is provided.

When the output image is transmitted, the user interface 116 according to an exemplary embodiment of the present invention outputs the image on the screen and receives and transmits a reference point of the parking position to be parked on the output screen. To this end, the user interface 116 according to the preferred embodiment of the present invention includes a liquid crystal display (LCD), a touch pad, a keyboard, and the like.

The sensing unit 120 includes a plurality of sensors, that is, a wheel speed sensor, a steering angle sensor, a yaw rate sensor, an acceleration sensor, and the like to detect a driving situation of a vehicle and assist the parking signal with an electric signal. Transfer to the position estimator 138 of the controller 130.

The parking assistance controller 130 includes a route plan generation unit 132, a route tracking control unit 134, and a position estimating unit 136 to park by using the target parking block transferred from the target parking position setting unit 110. Set a route plan for the vehicle, consider the location and driving conditions of the vehicle, track the set route plan, generate a control signal for controlling the vehicle to park in the target parking zone, and transmit the generated control signal to the active steering unit 140. .

The active steering unit 140 is a steering assistance device for inducing steering to obtain steering safety using various sensors and control devices with respect to a driver's steering input. The active steering unit 140 is provided with a control signal transmitted from the parking assistance control unit 130. Control the vehicle accordingly.

Here, the sensing unit 120, the parking assistance control unit 130 and the active steering unit 140 is the same or similar to the role and function of the conventional automatic parking system or semi-automatic parking system, a detailed description thereof will be omitted.

2 is a flowchart illustrating a parking block recognition method using hop transform according to a preferred embodiment of the present invention.

After photographing the parking space in the camera 112 to generate an input image and transmitting the same to the parking compartment recognition unit 114, the parking compartment recognition unit 114 converts the input image into a bird's eye view corner image (S210).

The parking compartment recognition unit 114 hop-transforms the bird's-eye view corner image to generate a hop space image (S220), and detects a plurality of peak pairs having the same direction at a predetermined distance from the hop space image to detect a plurality of peaks on the bird's-eye view corner image. The parking compartment display line is recognized (S230).

The parking section recognition unit 114 recognizing the plurality of parking section display lines recognizes the plurality of parking section display lines by checking whether there are corner pixels at the centers and boundaries of the plurality of parking section display lines on the bird's eye view corner image (S240). ).

The parking section recognition unit 114 that recognizes the plurality of parking section display segments, is closer to the camera 112 among the plurality of parking section display segments on the bird's eye view corner image, and is more vertically divided from the line of sight of the camera 112. Recognizing the guideline by selecting (S250).

The parking section recognition unit 114 that recognizes the guideline recognizes the portion where the difference in contrast is significantly smaller as the division display line by comparing the difference between the upper contrast and the outer contrast of the parking compartment display line along the guideline. S260).

The parking compartment recognition unit 114 recognizing the segmentation line segment recognizes the guideline and the segmentation line segment as the parking segment (S270).

Hereinafter, a process of recognizing a target parking section from an input image in the target parking position setting unit 110, in particular, the parking section recognition unit 114 according to an exemplary embodiment of the present invention will be described in detail.

1. Three coordinate systems

3 is an exemplary diagram for explaining three coordinate systems.

A fish eye lens is mounted on the camera 112, and a D (Distorted) coordinate system of an input image such as 3A is used to interpret an image photographed at a fixed height / fixed tilt angle using the fisheye lens. (Hereinafter referred to as 'D coordinate system'), U (Undistorted) coordinate system of the distortion correction image such as 3B (hereinafter referred to as 'U coordinate system'), bird's eye view B (Bird's Eye View) coordinate system (hereinafter referred to as 'B coordinate system') Three coordinate systems should be defined.

2. Fish Eye Lens Model

4 is an exemplary diagram for explaining a relationship between a D coordinate system and a U coordinate system.

Like 4A and 4B, the distance between the point Pd (Xd, Yd) on the D coordinate system and the image center Od (Oxd, Oyd) is called rd, and the point Pu (Xu, Yu) on the U coordinate system and the image center Ou (Oxu, If the distance between Oyu) is ru, rd can be expressed as ru's fifth order polynomial. Its inverse function can also be expressed as a fifth order polynomial. That is, ru may also be expressed as a fifth order polynomial of rd. Ignoring the tangential distortion between Pd and Pu, the one-to-one correspondence is established by the radial distortion modeled above.

3. Homography

5 is an exemplary diagram for describing a conversion process between a U coordinate system and a B coordinate system.

The point Pu on the U coordinate system and the point Pb on the B coordinate system correspond one-to-one by homography. Therefore, Bird's Eye View Transform T _B is defined as shown in Equation 1 by projecting an image plane on a plane as shown in FIG. 5.

6 is an exemplary diagram for describing a process of converting a pixel coordinate system into an image index coordinate system.

When the conversion K _j from the (x, y) coordinate system, which is the pixel coordinate system on the image plane as shown in FIG. 6, to the (j, j) coordinate system, which is the image index coordinate system, is defined in Equation 2 Is defined as:

7 is an exemplary diagram for describing a process of converting a world coordinate system to a bird's eye view index coordinate system.

A transformation K _B from a world coordinate system (x _w , z _w ) coordinate system as shown in FIG. 7 to a bird's eye view index coordinate system (j _B , i _B ) is defined as shown in Equation 3 below. .

If the homography H from the (j, i) coordinate system, which is the image index coordinate system, to the (j _B , i _B ) coordinate system, which is the bird's-eye view index coordinate system, is defined as K _I , T _B , K _B , Equation (5).

Therefore, Pb corresponding to Pu can be obtained by multiplying H, and Pu corresponding to Pb can be obtained by multiplying H ⁻¹ .

4. Edge Image of Bird's Eye View

An image of the corners of Bird's Eye View is created using the Sobel Edge Detector.

Table 1 shows the Sobel horizontal mask, and Table 2 shows the Sobel vertical mask.

Each pixel of the corner image is calculated as Equation 5 as the sum of the results of applying the Sobel Horizontal Mask and the result of applying the Sobel Vertical Mask.

Here, E is a pixel value of the corner image, B is a pixel value of the bird's eye view, and the final corner image is a result of binarization by applying a threshold to the E value.

8 is an exemplary view for explaining a view of a bird's eye view and a bird's eye view corner image.

When the input image of the D coordinate system is converted into the bird's-eye view corner image of the B coordinate system through the above-described process, it appears as the bird's-eye view shown in 8A and the bird's-eye view edge image shown in 8B.

5. Hope transform of bird's eye view corner image

Parking Lot Marking consists of Marking Line-Segments, each of which is represented by two parallel Line-Segments separated by a certain distance from the corner. The two parallel segments separated by a certain distance form a unique pattern on the hop space image. That is, a pair of peaks of similar height are formed having the same coordinates on the tilt (θ) axis and separated by the width of the display line segment on the distance (d) axis.

Hough Transform is the most common method used to find straight lines in binary images. One point (x, y) on the binary image is converted into (θ, d) which is a parameter of all the straight lines passing through this point. If all points on the binary image are transformed in this way and accumulated in the hop space with the linear parameters as axes, one straight line on the binary image forms one peak in the hop space. Accordingly, by performing peak detection on the hop space, a straight line on the binary image may be recognized.

Conventional hop transform uses the equation of cosθ · x + sinθ · y = ρ by using the inclination angles θ and ρ of the straight line as axes, but in the present invention, the characteristics of two parallel line segments are well represented by being separated by a certain distance. Use θ and d as axes. Here, d is the distance from the origin to the straight line multiplied by the sign of the intersection (Sign of Intersection). By using this to represent the equation of the straight line it can be defined as in Equation 6.

FIG. 9 is an exemplary diagram for describing converting a point on a bird's eye view corner image onto a hop space image.

Using a straight line equation, a point on the bird's eye view corner image such as 9A may be expressed as a curve on the hop space image as shown in 9B. 9A shows the point on the edge image on the x and y axes, and 9B shows the curve on the hop space on the θ and d axes.

If there are a plurality of segments on the aerial image, peaks corresponding to the segments appear on the hop space image. Since the height of the peak on the hop spatial image depends on the length of the line segment, it is impossible to select only valid peaks using a constant threshold. In addition, since the curve on the hop space image generated by one straight line merges with the curve produced by the other straight line, mutual interference and interference cannot be avoided.

Various studies have been conducted to solve this problem, and among them, many methods have been proposed in which a distribution around a peak on a hop space image is a butterfly pattern. It uses the shape of a butterfly to detect more robust peaks, and finds the start point, endpoint, and thickness of a line segment. However, it is not suitable for real-time applications because it requires a large amount of computation and time to analyze the surrounding distribution for all points on the hop space image.

10 is an exemplary diagram showing a straight line on the bird's eye view corner image as a plurality of curves on the hop space image.

10A shows a straight line on the bird's eye view corner image, 10B shows a straight line on the bird's eye view corner image as a curve on the hop space image, and 10C shows a three-dimensional butterfly shape curve on the hop space image. Here, the θ axis is represented by a resolution of 0.2 °, the range is -90 ° to 90 °, the number represents the index. In addition, the d-axis is represented by the resolution of 1 pixel, the range is -1000 to 1000, the number represents the index.

In the present invention, it is assumed that the width of the display line segment of the parking partition display is substantially constant, and the display line segment is found by finding two parallel line segments separated by a predetermined distance on the corner image.

11 is an exemplary diagram showing two parallel straight lines on a bird's eye view corner image on a hop space image.

11A shows two parallel straight lines on the bird's eye view corner image, 11B shows two parallel straight lines as multiple curves on the hop space image, 11C shows an enlarged part of the hop space image, and 11D shows a hop space. Multiple curves on the image will be presented in three dimensions of a butterfly shape. Here, the θ axis is represented by a resolution of 0.2 °, the range is -90 ° to 90 °, the number represents the index. In addition, the d-axis is represented by the resolution of 1 pixel, the range is -1000 to 1000, the number represents the index.

As shown in FIG. 11, the hop transform results of two parallel line segments separated by a certain distance W form a characteristic peak pair on the hop space image. The two peaks corresponding to each line segment have the same θ value and are separated by W on the d-axis. In addition, the two peaks are about the same size, and a deep valley is formed between the two peaks. When such a feature is applied to the bird's eye view of the parking lot, it may appear as shown in FIG. 12.

12 is an exemplary view showing several parallel straight lines of the parking lot bird's eye view on the hop space image.

When the hop transform result of the binary edge image of the bird's eye view of the parking lot as shown in 8B of FIG. 8 is displayed, it may be represented as shown in FIG. 12. That is, it can be confirmed that there is a peak pair having the above-described feature for every display line of the parking lot such as 12A. In this case, since the parking segment display segment, which is the characteristic to be searched, shows a very characteristic pattern on the hop space image, it is more effective to recognize the peak pair directly than to recognize the straight line in the binary edge image by the conventional hop transform. It can be seen that even in the efficient. By looking at the hop space of a binary edge image in three dimensions, one can predict how difficult it will be to find each peak.

6. Peak Pair Detection on Hope Spatial Image

The peak pairs on the hop space having the same θ value on the hop space and separated by a predetermined distance W from the d axis can be found by one-dimensional filtering and clustering. Since the wing width and the starting point of the butterfly pattern, which means the dispersion around the peak, depend on the length and position of the line segment, the general rule of the θ-axis direction for the peak pair to find is not found.

Therefore, candidate points of the peak pair are detected by performing one-dimensional filtering on the peak pair in the hop space in the d-axis direction, and clusters are searched through morphological dilation and connected component search. Detect. Through this, it can be seen that the center of gravity of the cluster is an excellent estimate of the marking line of the parking section marking corresponding to the peak pair.

13 is an exemplary view for explaining a peak pair detection process according to a preferred embodiment of the present invention.

Referring to FIG. 13, an equation for detecting peak pairs on a hop spatial image may be derived as shown in Equation 7 below.

14 is an exemplary diagram for describing a method of detecting peak pairs on a hop spatial image according to a preferred embodiment of the present invention.

Here, HS (θ, d) is the sum of the contributions of the pixels on the corner image with respect to a straight line with θ and d as parameters of the hop space element. When the inspection target coordinate in the hop space is (θ, d), HS (θ, d), HS (θ, dW), and HS (θ, d + W) are valleys and HS (θ, dW / 2) It can be seen that (θ, d) is the display line when HS (θ, d + W / 2) is the peak. When the HS (θ, d) values of the peak inspection positions are P _A and P _B , respectively, the maximum value is determined within a predetermined range by considering that the displacement between the inspection target coordinate and the peak depends on the width of the display line segment. Robustness of the detection was ensured.

S (θ, d) is a value obtained by applying one-dimensional filtering in the d-axis direction, and similar L (θ, d) is defined as a value between 0 and 1 only when S (θ, d) is positive. Since S (θ, d) has a maximum value of P _A + P _B when the valleys are all zero, dividing S (θ, d) by P _A + P _B is intended to generalize between 0 and 1. . Here, multiplying the ratio of the minimum and maximum values of P _A + P _B is to evaluate the weighted values when the two peaks have the same size. When L (θ, d) thus obtained is larger than the threshold θ _{dual peak} , one point θ, d on the hop space becomes a candidate for the display line.

As described above, the predicted display line candidates form a cluster on the hop space image, but some candidates may not be connected to the cluster according to the thickness and length of the edge segment. In order to compensate for such a disadvantage of binarization, when a cluster is detected by performing a dilation using a 5 × 5 square kernel, the cluster's center of gravity is a parameter of the display line.

14 is an exemplary view for explaining a process of extracting a parking section display line by detecting a cluster.

14A shows the cluster candidates detected on the hop space image, 14B shows the center of gravity of the cluster candidates on the hop space image, 14C shows the parking block line on the bird's eye view corner image estimated by the center of gravity of the cluster candidates, 14D expresses the parking segment display line estimated on the hop space image in D coordinate system.

7. Recognition of Marking Line-Segment

A section that satisfies the condition of the display line segment is detected on the detected parking section display line, and the parking section display line segment is recognized. In this case, the condition of the display line segment is the same as prior knowledge used to find the peak pair in the hop space. That is, the display line segment is composed of two edge segments parallel to each other by a predetermined distance W.

FIG. 15 is an exemplary diagram for describing a process of obtaining an end point of a parking segment display line from a bird's eye view corner image.

To find the two endpoints within the range of the bird's eye view corner image, as shown in 15A, the point closer to the vehicle's camera is called P ₀ (x ₀ , y ₀ ) and the farther point is P ₁ (x ₁ , y ₁ ), it is checked from the P ₀ to P ₁ direction to see if the condition of the display line is satisfied.

As shown in 15B, the likelihood of the point P (s) on the line segment P ₀ P ₁ separated by P ₀ by s is examined by checking whether there is no corner pixel at the center of the parking zone indicator line and corner pixel at the boundary of the display line. Evaluate L (s) with values ranging from 0 to 1. The unit vector d in the P ₀ P ₁ direction can be obtained as shown in Equation 8 from the equation of the straight line. The unit vector n perpendicular to d can be obtained by rotating d by 90 °. Points on the s as off line from P ₀ P ₀ P ₁ P (s) is d and a can be obtained by adding the product of the distance s, here n, and the product of the vertical distance w by adding the reference mark boundary point B (s, w) in the P ₀ Can be obtained. When examining the presence of edge pixels at the center and the display line boundary, several pixels are examined in the vertical direction to increase noise robustness.

16 is a flowchart for explaining a process of recognizing parking segment display line segments.

By using hysteresis in applying a threshold value to L (s) to determine the _start point P _start and the end point P _stop of the display line segment, the noise robustness can be improved. When determining the start point, L (s) must be continuously greater than the threshold θ _{start check} for W / 2, and when determining the end point, L (s) is continuously smaller than the threshold θ _{stop check} for W / 2. Through this, it is possible to prevent the change during the short interval of L (s) to be mistaken for the starting point and the end point. At this time, since the camera searches in the direction of P ₁ from the point P ₀ close to the camera, P _start is set closer to the camera than P _stop .

17 is an exemplary view illustrating a recognition result of a parking section display line segment on a bird's eye view.

As shown in FIG. 17, the result of recognizing the parking segment display segment on the bird's eye view corner image as shown in FIG. 16 is shown in the bird's eye view. Here, the red dot represents the start point and the blue dot represents the end point.

8. Guidelines Recognition

As described above, the guideline may be recognized by selecting a line segment closer to the distance from the camera 112 and more perpendicular to the line of sight direction among the recognized parking segment display line segments. The guideline is a display line segment representing the boundary between the parking lot and the road among the display line segments forming the parking compartment mark.

The distance from the point to the line segment is either the minimum value of the distance from the point to the line or the distance from the point to the end point, depending on the location of the foot of perpendicular. That is, if the foot of the repair line is within the line segment, the distance from the point to the line is the distance from the point to the repair line. If the foot of the repair line is outside the line segment, the smaller value is selected from the distance from the point to both end points. The repair foot P _C (X _C , Y _C ) is defined as the intersection of the straight line perpendicular to the line segment S and the extension line (y = ax + b) passing through the point P (X _P , Y _P ). When the end points of the line segments are E ₁ and E ₂ , it can be determined by the sign of the dot product of the vectors P _C E ₁ and P _C E ₂ whether the feet P _C of the repair line are between the _two ends. In other words, if the sign of the dot product is negative, the foot of the waterline is between both ends, and if the sign of the dot product is positive, the foot of the waterline is outside both ends. This may be represented by Equations 9 and 18 as shown in Equations and drawings.

18 is an exemplary diagram illustrating a vector for recognizing a guideline. 18A shows the case where the foot of the repair is within the line segment, and 18B shows the case where the foot of the repair is outside the line segment.

In the case of the parking segment display segment, since the end point close to the camera is defined as P _Stsrt and the far end point is defined as P _Stop , Equation 9 can be simplified as shown in Equation 10.

19 is an exemplary diagram for describing a process of stably detecting a guideline regardless of a detection position of an end point of a display line segment.

The guideline can be stably detected regardless of the detection position of the end point of the display line by correcting the distance to the display line using the fact that the guide line is likely to be perpendicular to the direction of the camera line of sight. Display segments The extent directed from the camera to the line segment point P _start unit vectors u (P, P _start) and a unit vector directed to the end point P _stop at a time point P _start of the segment u (P _start, P _stop) perpendicular with respect to the viewing direction Evaluate internally. Since the dot product becomes smaller as the line segment is perpendicular to the line of sight, the smaller the product of the dot product with the distance from the camera to the display line segment, the more likely it is to be a guideline. The modified distance D (P, S) from the point P to the line segment S is defined as follows, and the guideline is determined as the display line segment with the smallest modified distance. In the above figure, even though the viewpoint of the partition line segment S ₂ is closer than the viewpoint of the guideline S ₁ , it has a direction similar to the visual direction, and thus has a larger correction distance than the guideline.

20 is an exemplary view illustrating a result of recognizing a guideline according to a preferred embodiment of the present invention.

As described above, the equation is defined as in Equation 11, and the result of recognizing the guideline is shown in FIG.

9. Recognition of Dividing Marking Line-Segment

According to the guideline, the difference between the intensity above the parking compartment marking line and the intensity of the outside light can be used to recognize the segmentation line segment of the parking compartment mark. Parking zone markings consist of guidelines that separate parking zones and roads, and vertical segmentation lines. By detecting a portion where the contrast and the outer contrast of the parking compartment mark segment are significantly smaller than the surroundings, the 'T'-Junction of the guideline and the segment mark segment can be successfully recognized. In particular, by detecting not only the parking division mark line of the guideline but also the entire extension line of the guideline, it is possible to recognize an unclear (Blur) dividing line far from the camera 112, which could not be recognized as a peak pair on the hop space image. have. Contrast I _ON (s) and Outer Contrast I _OFF (above the parking segment indicator line at the point where the extension line of the guideline meets the bird's eye view boundary and the camera 112 opposite the unit vector d in the guideline direction s). s) is defined as in Equation 12.

 In this case, n is a unit vector in a direction away from the camera 112 while being perpendicular to d, and W represents the width of the parking compartment display line segment. B (P) means the contrast of the point P (x, y) on the bird's eye view.

21 is an exemplary diagram for describing a process of recognizing a segment display line segment according to an exemplary embodiment of the present invention.

As described above, the division display line segment can be recognized as shown in FIG. 21 by using the difference between the upper contrast and the outer contrast of the parking compartment display segment of the guideline.

If the equation shown in Equation 12 is represented as a graph, it may be shown as shown in FIG.

22 is an exemplary view showing guidelines extension lines and I _ON and I _OFF of a bird's eye view.

The ratio of the difference between I _ON (s) and I _OFF (s) for I _ON (s) is called L _dividing (s), and when L _dividing (s) is less than the threshold θ _dividing longer than W / 2 It is recognized as a segmentation line segment. Equation 13 shows the generalization of I _ON (s) and I _OFF (s) by _ividing (s).

In general, the lighting brightness is not constant, so the values of I _ON and I _OFF vary greatly, but it can be seen that I _ON (s) and I _OFF (s) change in a similar degree at a specific s. Therefore, by dividing the difference between the two values by I _ON (s), it is possible to normalize to a value between 0 and 1, and at the same time remove the influence of the illumination brightness.

FIG. 23 is an exemplary diagram showing L _dividing (s) and a threshold value, and FIG. 24 is an exemplary diagram showing a recognition result of segmented display lines.

10. Occlusion Impact Analysis

The parking compartment recognition method according to an exemplary embodiment of the present invention can be successfully performed even in a situation in which cars are parked next to an empty parking compartment.

25 is an exemplary view illustrating a result of recognizing a parking compartment according to a parking compartment recognition method according to a preferred embodiment of the present invention.

Obstacles including sunshade cause a severe distortion in a radial manner in the bird's eye view, as shown in FIG. 25. 25A shows an obstacle on the distortion corrected image. 25B shows distortion caused by obstacles on the bird's eye view. 25C shows the distortion removed by detecting peak pairs on the aerial view edge image. The 25D shows guidelines that were successfully recognized despite the parking segment markers that were incorrectly recognized in the bird's eye view. 25E shows the result of the calculation of L _dividing . 25F shows the result of recognizing the additional segmentation segment on the bird's eye view.

The bird's eye view is created on the premise of Plane Surface Constraints, because the obstacles do not meet these prerequisites. That is, obstacles which do not belong to the floor surface are projected as drawn on the distant floor. In particular, the light objects on the dark background elongate, creating a distortion that can be confused with the parking segment marking line. In the parking lot recognition method according to the preferred embodiment of the present invention, such distortion is eliminated in two steps.

1) On the bird's eye view corner image, the parking segment indicator segments appear as two parallel edges separated by a certain distance. Therefore, the distortions that do not satisfy this are removed in the peak pair recognition process and the parking segment display segment recognition process on the hop space image.

2) If a car is normally parked next to an empty parking section, the distortion caused by the vehicle's bird's eye view is farther away than the guideline. Therefore, the parking compartment recognition method according to the preferred embodiment of the present invention can successfully recognize the guidelines despite the distortion of the vehicle. Once the guideline has been successfully recognized, other noise can be removed by recognizing the segmentation line connected by the letter 'T' in the guideline.

In particular, since the guideline tends to appear as long and sharp edges even when the lighting is dark and many corners of the parking lot indicator line are not detected, the guideline can be recognized successfully. Once the guideline is recognized, it is possible to recognize the parking compartment mark successfully by recognizing the segmentation line segment not found in the recognition of the parking compartment mark line.

FIG. 26 is an exemplary view illustrating a process of recognizing segment display line segments in backlight. FIG.

26A shows the back light in the distortion correction image. 26B shows a case where an edge is detected only in a part of the parking mark on the bird's eye view corner image. 26C shows the result of the guideline recognition on the bird's eye view. 26D shows the recognition of the segmented line segment in addition to the recognition result of the parking section on the bird's eye view.

27 is an exemplary diagram illustrating a process of recognizing a segment display line segment when light is reflected.

27A shows the reflection of light on the floor in the distortion correction image. 27B shows the detection of peak pairs when a large number of false peak pairs are found due to light reflection and side difference on the bird's eye view. The 27C shows the removal of the false recognition by recognizing the parking compartment segment on the bird's eye view. In the bird's eye view, the 27D removes false recognition by recognizing the guideline and the separator line.

The above description is merely illustrative of the technical idea of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and changes without departing from the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention but to describe the present invention, and the scope of the technical spirit of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

As described above, according to the present invention, by recognizing the parking section using the image of the camera in the vehicle, it is possible to accurately and conveniently set the target position to be parked by the driver.

In addition, the parking compartment display line can be accurately recognized by detecting the peak pairs on the hop space having the same inclination by being separated by a certain distance by utilizing the point where the parking compartment displayed in the parking space has a constant width.

In addition, the parking section display lines closer to the camera and more perpendicular to the direction of the camera's line of sight as the guide lines may be recognized as guide lines, thereby recognizing the guidelines having a great influence on the parking section recognition.

In addition, the parking compartment can be accurately recognized by recognizing the segmentation line segment by contrasting the upper and the outer contrast of the guideline along the guideline.

In addition, by distinguishing the contrast line not only the guide line but also the contrast line, the segment display line segment of the blurry part of the camera image from a remote location can be accurately recognized.

In addition, even when an obstacle is present in the parking space, distortion due to the obstacle may be minimized through the process of detecting the peak pair on the hop space image and recognizing the division display segment.

In addition, even when parking compartment is difficult due to backlight, light reflection, etc., incorrect recognition may be eliminated by a peak pair detection method, a guideline recognition method, a division mark line recognition method, and the like, thereby accurately recognizing the parking compartment.

Claims (16)

An apparatus for recognizing a parking compartment by including a camera for photographing a parking space in a vehicle and generating an input image, and a user interface for receiving and transmitting a reference point to be parked among the parking spaces from a driver and a driver. When the input image is received from the camera, the input image is converted into a bird's eye view corner image, and the bird's eye view corner image is transformed by Hough to generate a hop space image, and then separated by a predetermined distance from the hop space image. Recognizing a plurality of parking section display lines on the bird's eye view corner image by detecting a peak pair having a direction, recognizing a plurality of parking section display lines on the plurality of parking section display lines, and the plurality of parking section display lines After recognizing the guideline, the parking section recognition unit for recognizing the guide line and the plurality of display line segments by using the guide line to recognize the plurality of display line segments. Parking compartment recognition apparatus using a hop transform comprising a. According to claim 1, wherein the parking section recognition unit, The peak pairs are one-dimensionally filtered along the distance axis on the hop spatial image to detect a plurality of peak pair candidates spaced apart by a predetermined distance, and the morphological dilation and linkage elements in the plurality of peak pair candidates. Connected Component) After detecting the cluster (Cluster), the parking compartment recognition apparatus using a hop transform, characterized in that for recognizing the center of gravity of the cluster as the peak pair. The parking compartment recognition unit of claim 2, And a plurality of peak pair candidates are detected by obtaining a maximum value within a predetermined range centering on a target coordinate of the hop space image. The method of claim 3, wherein the parking section recognition unit, And a plurality of peak pair candidates are detected by weighting two peaks having a maximum value of the same size with each other. The method of claim 2, wherein the parking section recognition unit, A parking compartment recognition apparatus using a hop transform, characterized in that for detecting the cluster by performing a dilation by using a kernel having a rectangular shape of a constant size. According to claim 1, wherein the parking section recognition unit, After obtaining two end points of the parking segment display line in the bird's eye view corner image, whether two parallel edge segments exist parallel to each other by a predetermined distance, proceeding from the point closer to the camera among the two end points. The parking compartment recognition apparatus using the hop transform, characterized in that for recognizing the parking compartment display line segment. The method of claim 6, wherein the parking section recognition unit, Parking segment recognition using a hop transform characterized in that there is no corner pixel at the center between the parking segment display lines and the presence of corner pixels at the boundary of the parking section display line. Device. According to claim 1, wherein the parking section recognition unit, A parking segment recognition device using hop transform, characterized in that the parking section display line segment closest to the camera from among the plurality of parking segment display line segments and more perpendicular to the direction of the line of sight from the camera is recognized as the guide line. . The method of claim 8, wherein the parking section recognition unit, The guide section shows a parking segment display line segment having a smaller product of a unit vector directed from the camera to a viewpoint of the plurality of parking segment display lines and a dot product of a unit vector pointing to an end point at the time points of the plurality of parking segment display lines. Parking compartment recognition apparatus using hop transform, characterized in that the recognition as a line. According to claim 1, wherein the parking section recognition unit, Recognition of the parking compartment using the hop transform, characterized in that the portion of the difference between the intensity (Intensity) above the guideline and the contrast of the outside of the guide line becomes smaller than the threshold along the guideline Device. The method of claim 10, wherein the parking section recognition unit, A parking compartment recognition device using a hop transform, characterized in that not only the parking compartment display line segment corresponding to the guideline but also the portion where the difference in contrast between the entire extension line of the guideline becomes smaller than a threshold is recognized as the segmentation display segment. . The method of claim 10, wherein the parking section recognition unit, A parking compartment recognition device using a hop transform, wherein the division mark line segment is recognized by dividing a difference between a contrast value above the guideline and a contrast value outside the guideline by the contrast value of the guideline. . In the parking compartment recognition unit that is connected to the user interface unit for receiving and transmitting a reference point to be parked in the parking space from the camera and the driver to generate the input image by shooting the parking space in the vehicle and transmits the, In the method of recognizing the parking compartment, (a) receiving the input image to generate a bird's eye view corner image, and hop transforming the bird's eye view corner image to generate a hop space image; (b) detecting a plurality of peak pairs having the same direction at a predetermined distance from the hop spatial image; (c) converting the hop spatial image into the bird's eye view corner image to recognize the plurality of peak pairs as a plurality of parking section display lines; (d) recognizing a plurality of parking section display lines from the plurality of parking section display lines; (e) recognizing a guideline from the plurality of parking section indicator lines; And (f) recognizing a plurality of segmentation line segments using the guideline Parking compartment recognition method using a hop transform, characterized in that it comprises a. The method of claim 13, wherein the parking section recognition method, After the step (f), (g) recognizing the guide line and the plurality of division mark segments as the parking compartment; And (h) receiving the reference point and projecting the reference point to the guidelines to establish a target parking compartment; Parking segment recognition method using a hop transform, characterized in that it further comprises. The method of claim 13, wherein step (b) comprises: (b1) detecting a plurality of peak pair candidates through one-dimensional filtering to detect peak pairs separated by the distance in a distance axis direction on the hop spatial image; (b2) detecting clusters by performing morphological dilation and connected component search on the plurality of peak pair candidates, and detecting the center of gravity of the cluster as the plurality of peak pairs; step Parking compartment recognition method using a hop transform, characterized in that it comprises a. In the parking assistance system for assisting the parking of the vehicle by recognizing the parking compartment When the input image is received from the camera, the input image is converted into a bird's eye view corner image, and the bird's eye view corner image is transformed by Hough to generate a hop space image. Recognizing a plurality of parking segment display lines on the bird's eye view corner image by detecting a peak pair, extracting guideline and a plurality of segment display lines from the plurality of parking segment display lines, and recognizing them as the parking segment. A target parking position setting unit for converting the bird's eye view corner image into an output image and outputting a reference point from a driver, projecting the reference point to the guideline, and setting and delivering a target parking compartment; A sensing unit having a plurality of sensors for recognizing a driving state of the vehicle to generate and transmit driving state information of the vehicle; Receive the information on the driving situation from the detection unit to estimate the location of the vehicle, generate a route plan for parking the parking section by receiving the target parking section and parking in the parking section in consideration of the location Parking assistance control unit for generating and transmitting a control signal for the; And An active steering unit which drives the vehicle according to the control signal when receiving the control signal Parking assistance system using a hop conversion, characterized in that it comprises a.

Images